In previous decades, various methods have been proposed and constructed to emulate human-like behavior, many which were bio-mimetic. That is, they were suggested by the underlying human biological elements of the human brain. While these have been successful in part, they have failed to permit accurate emulation of the brain on a large scale. Some bio-inspired concepts such as fuzzy logic have made relatively few inroads into commercial markets. Fuzzy logic and rules-based applications have been very niche-like and limited, although within those niches, fuzzy logic has performed quite well. Neither have proven amenable to implementation on brain-level scales, however.
Throughout this document, “brain emulation” and “brain model” have been used interchangeably as needed to best convey intent.
Some approaches of a prior system have depended upon bio-inspired neural networks, such as shown in FIG. 1.
In a neural network, for example, a set of neurons 1 and 2 are assumed to be stimulated by some external means 95, each neuron typically representing a fact. It fires in proportion to the present state of recognition of that fact. They may be connected to other neurons 3 and 4, with the connections implying a specific relationship between them. Interposed in the connections between the neurons may be a set of weights 5 which control the influence of the ‘input’ neurons 1 and 2 over the ‘output’ neurons 3.
Finally, various forms of control such as inhibitors may be implemented, such as for neuron 4. In this case, the firing on neuron 2 may inhibit the influence of upper neuron 13 upon output neuron 4, as indicated by inhibitor 6. The implementation may either be within neuron 4 or may precede it. The organization and interconnect of the network dictates that certain present input conditions 95 will yield the desired output results 96.
Another popular form of a prior system involves the use of a system of rules, such as depicted in FIG. 2. In this case, a set of input conditions 95 are acted upon by a set of rules 7, to produce a set of pre-defined desired output results 96. The rules translate the set of input conditions into a set of signals that depicts the desired results. A set of feedback paths 97 are further compared with the inputs 95 to modify the rules 7, permitting the outputs 96 to converge on the desired results. These are typically referred to as “first principles” systems.
While rules-based systems tend to produce the desired results, they can be very complex and require enormous amounts of computing power. Many hundreds of thousands of rules may be generated, and the results are not always accurate. Typical applications such as text-to-speech that use rules-based mechanisms do not have good accuracies unless extensive operator-specific training is used. Inaccurate results and excessive computational power have prevented pervasive applications of rules-based systems.